A Latent Class Modeling Approach for Differentially Private Synthetic Data for Contingency Tables

We present an approach to construct differentially private synthetic data for contingency tables. The algorithm achieves privacy by adding noise selected summary counts, e.g., two-way margins of the table, via Geometric mechanism. posit underlying latent class model estimate parameters based on noisy and generate using estimated model. This allows agency create multiple imputations with no additional loss, thereby facilitating estimation uncertainty in downstream analyses. illustrate a subset 2016 American Community Survey Public Use Microdata Sets.